Method get_loras

scripts/trt.py:198–240

(self, p)

        return best, best_hr

    def get_loras(self, p):
        lora_pathes = []
        lora_scales = []

        # get lora from prompt
        _prompt = p.prompt
        extra_networks = re.findall("\<(.*?)\>", _prompt)
        loras = [net for net in extra_networks if net.startswith("lora")]

        # Avoid that extra networks will be loaded
        for lora in loras:
            _prompt = _prompt.replace(f"<{lora}>", "")
        p.prompt = _prompt

        # check if lora config has changes
        if self.lora_hash != "".join(loras):
            self.lora_hash = "".join(loras)
            self.update_lora = True
            if self.lora_hash == "":
                self.lora_refit_dict = {}
                return
        else:
            return

        # Get pathes
        print("Apllying LoRAs: " + str(loras))
        available = modelmanager.available_loras()
        for lora in loras:
            lora_name, lora_scale = lora.split(":")[1:]
            lora_scales.append(float(lora_scale))
            if lora_name not in available:
                raise Exception(
                    f"Please export the LoRA checkpoint {lora_name} first from the TensorRT LoRA tab"
                )
            lora_pathes.append(
                available[lora_name]
            )

        # Merge lora refit dicts
        base_name, base_path = modelmanager.get_onnx_path(p.sd_model_name)
        refit_dict = apply_loras(base_path, lora_pathes, lora_scales)

        self.lora_refit_dict = refit_dict

    def process(self, p, *args):
        # before unet_init